· Early Childhood Teacher
I asked an AI assistant for sustainability theme angles for my 3-4 year olds, and it handed me five in about ten seconds. Water play. Recycling craft. A garden patch. An animal habitat. A weather chart. All sensible. All the kind of thing you’d nod at in a planning meeting. I picked the garden patch, and not because it read best. I picked it because our service had just put a raised bed in at the back of the yard. This post is about that choice. Using AI for lesson planning in early childhood education to generate options, then letting the room and the yard decide which one was actually real.
Why I turned to AI for sustainability themes (and the prompt I used)
Planning for 3-4 year olds eats time in the parts of the day you don’t have. I get maybe twenty minutes during rest time. Half of that goes to documentation I’m already behind on. So when our room committed to a sustainability focus for the term, I didn’t want to stare at a blank programming sheet. I wanted a list to react to.
That’s the honest use case here. I wasn’t asking the model to write my curriculum. I was asking it to break the blank-page freeze so my own judgement had something to push against.
The prompt I used was specific: “Give me five sustainability curriculum theme angles for a kindergarten room of 3-4 year olds in an Australian long day care service. Keep each one play-based and hands-on. One sentence each.” Place-specific, age-specific, short. I’ve learned that a vague prompt gives you a vague list, and a vague list wastes the twenty minutes I was trying to save. If you want a deeper walkthrough of prompt shape, I’ve written about that in how I iterated four prompts to make AI sound like me.
The five ideas the AI gave me
Here’s what came back, near enough verbatim:
- Water play: exploring how we use and save water at the trough.
- Recycling craft: sorting and making with clean recyclables.
- Garden patch: planting, watering, and watching food grow over the term.
- Animal habitat: building homes for local bugs and birds.
- Weather chart: tracking daily weather and talking about seasons.
Read them on a screen and every one looks fine. That’s the trap. On paper they’re equally valid, equally “alignable” to a learning framework, equally easy to write up in a program. The model has no way of knowing which of them my room can physically do this week, because the model has never stood in my yard.
Four of those five would have been fine on paper and dead in practice. The recycling craft needs a steady stream of clean materials I don’t reliably have. The animal habitat needs a corner of garden we hadn’t cleared. The weather chart is a five-minute mat ritual, not a term-long inquiry. Water play I could do, but it doesn’t grow. There’s nothing for a child to come back to on Thursday and find changed since Monday.
The filter: does it live in our actual environment?
Here’s the filter I ran them through. It has nothing to do with how good the idea sounds. Three questions. Touchable: a child can put their hands on the real thing, not a photo of it. Revisitable: it’ll still be there next week, visibly different, and the children can notice the change. Child-led: a three-year-old can drive it, not just an adult running the show. That’s the whole test.
The raised garden bed passed all three. The children could press seeds in with their own fingers. They could fill the watering cans and tip them out, badly, gloriously, every morning. And across the term something would actually change. A seed becomes a green thread, a green thread becomes a leaf, a leaf becomes a bean they can pick. That arc is the whole point.
This is what the SaaS planning blogs miss. They stop at “AI generates plans, here are the risks.” Nobody shows the selection step, which is the part where the educator’s knowledge of her own four walls and her own yard does the real work. The AI gave me five doors. Standing in the room told me four of them opened onto nothing.
Building the garden-patch theme into real planning
Once I’d chosen, the framework mapping was the easy part. And notice it came after the place decision, not before it. Sustainability is now a named principle in the Early Years Learning Framework V2.0, spanning environmental, social and economic dimensions, and a garden patch sits naturally across all three: caring for living things, sharing the watering roster fairly, and not wasting what we grow. I didn’t have to force the alignment. The bed did it for me.
The pacing, though, I handed to the yard, not to a plan. Some weeks nothing visible happened and the children lost interest, so we went and looked at worms instead. One week a possum got the seedlings overnight and that became three days of conversation I never programmed. I leaned on play-based, child-led practice here on purpose; the NAEYC position statement on developmentally appropriate practice frames learning as something built through active engagement, play and inquiry, not delivery against a fixed script. The garden let me actually work that way. The weather chart never would have.
In my practice, the documentation got easier too. Real photos of real hands in real soil beat any clip-art activity sheet, and the children’s own questions wrote half my observations for me. If you’re weighing where this kind of tooling helps versus where it quietly hurts your programming, I’ve laid that out in writing child observations with AI without losing your voice.
Where the AI was wrong about my room (an opinion)
I’ll say this plainly, because the sector is being sold a softer version of it: AI is good at generating curriculum ideas and genuinely bad at choosing them. The choosing is the actual teaching. The model produced five clean options in seconds. It could not tell me that four of them had no home in my space, because it has no space. It doesn’t know we’d just built a raised bed. The unreliable recycling stream, the overgrown bug corner, water play that doesn’t change across a term — none of that is on a screen for it to read. Place-knowledge is exactly the thing these tools can’t have. And place-knowledge is most of the job.
So I deleted four ideas and kept one, and I’d argue the deleting was worth more than the generating. Anyone can produce a list. Knowing which item on it your three-year-olds can touch on a wet Tuesday in your yard — that’s the part no model gets to do for you, and in my view we should stop pretending it might. Ask your supervisor or educational leader how they’d want this documented, but own the selection yourself.
How to run this filter on your own AI ideas
Forget my exact prompt. What matters is the gate you run after the AI answers, before anything goes near your program. Four questions:
- Can a child touch the real thing this week? Not a printout of it. The actual thing.
- Will it still be here next week, visibly changed? If it can’t be revisited, it can’t carry a term.
- Can a three-year-old lead it? If it only works adult-run, it’s an activity, not an inquiry.
- Does it already live in our space, or am I building it from scratch? Existing beats aspirational nearly every time.
Run those four and the slick options usually fall away, and you’re left with the one your environment was already half-ready to hold.
TL;DR / Key takeaways
- I used AI for lesson planning in early childhood education to generate five sustainability themes for 3-4s, then chose with my own judgement — the AI brainstorms, the educator selects.
- I picked the garden patch because it was touchable, revisitable and child-led, and it already lived in our yard as a new raised bed.
- The other four read fine on screen but had no real home in our space this term, so they’d have been dead in practice.
- Framework alignment (EYLF V2.0 sustainability principle) came after the place decision, not before it.
- Use a four-question filter on any AI idea: touchable, revisitable, child-led, already-present. The choosing is the teaching.
Last reviewed: 3 June 2026.
Sources
- ACECQA — Information Sheet: EYLF Principles — Sustainability
- NAEYC — Developmentally Appropriate Practice (DAP) Position Statement
Fact-checked 2026-06-03. Last reviewed 2026-06-03.